Model-based adaptive segmentation

Abstract

A machine vision method is provided for segmenting an image based on an adaptive thresholding approach that exploits prior knowledge of the characteristics, such as area, of the object to be segmented. Adaptation to varying gray-values of the candidate object/background from scene-to-scene is obtained by using a sampling window which provides gray-value statistics of the candidate object to be segmented. Next, using these statistics, a characteristic of the image of the object to be segmented, such as its area or its gray-value standard deviation, and the histogram of the image, the segmentation is performed. The invention is especially useful for detecting the presence/absence of adhesive on printed circuit boards, where both the appearance of the printed circuit board and the solder paste commonly found thereon make segmentation difficult. In case of presence/absence determination, an additional step of shape matching is employed.

What is claimed is: 
     
       1. A method for segmenting a run-time image having at least a background region and a candidate object region, the method comprising the steps of: acquiring knowledge of a characteristic of a model object region in a model image;   constructing a sampling window for acquiring gray-value statistics on a candidate object region in said run-time image;   acquiring gray-value statistics on a candidate object region in said run-time image;   creating a gray-value histogram based on said run-time image;   using said gray-value statistics on said candidate object region, and said knowledge of said characteristic of said model object region, to determine a high threshold and a low threshold;   comparing each pixel of said run-time image with at least one of said high threshold and said low threshold to provide a binary comparison image including a plurality of binary comparison values, each binary comparison value indicating whether each said pixel corresponds to said candidate object region.   
     
     
       2. The method of claim 1, said characteristic of said model object region being the area of said model object region. 
     
     
       3. The method of claim 1, said characteristic of said model object region being the standard deviation of the gray-value of pixels within said model object region. 
     
     
       4. The method of claim 1, further including the steps of: performing connectivity analysis on said binary comparison image to provide run-time geometric object features;   performing connectivity analysis on a train-time binary comparison image to provide train-time geometric object features; and   comparing said run-time geometric object features with said train-time geometric object features to determine presence/absence of said object.   
     
     
       5. The method of claim 1, further including the steps of: using template matching to compare said binary comparison image with a train-time comparison image so as to determine presence/absence of said object.   
     
     
       6. An apparatus for segmenting a run-time image having at least a background region and a candidate object region, the method comprising the steps of: means for acquiring knowledge of a characteristic of a model object region in a model image;   means for constructing a sampling window for acquiring gray-value statistics on a candidate object region in said run-time image;   means for acquiring gray-value statistics on a candidate object region in said run-time image;   means for creating a gray-value histogram based on said run-time image;   means for using said gray-value statistics on said candidate object region, and said knowledge of said characteristic of said model object region, to determine a high threshold and a low threshold;   means for comparing each pixel of said run-time image with at least one of said high threshold and said low threshold to provide a binary comparison image including a plurality of binary comparison values, each binary comparison value indicating whether each said pixel corresponds to said candidate object region.   
     
     
       7. The apparatus of claim 6, said characteristic of said model object region being the area of said model object region. 
     
     
       8. The apparatus of claim 6, said characteristic of said model object region being the standard deviation of the gray-value of pixels within said model object region. 
     
     
       9. The apparatus of claim 6, further including the steps of: means for performing connectivity analysis on said binary comparison image to provide run-time geometric object features;   means for performing connectivity analysis on a train-time binary comparison image to provide train-time geometric object features; and   means for comparing said run-time geometric object features with said train-time geometric object features to determine presence/absence of said object.   
     
     
       10. The apparatus of claim 6, further including the steps of: means for using template matching to compare said binary comparison image with a train-time comparison image so as to determine presence/absence of said object.